Apple said the speech recognition models that power the voice-to-text feature might show words with some phonetic overlap.